topic Re: Proc phreg /interaction between non-time covariates in Statistical Procedures

Proc phreg /interaction between non-time covariates

Cruise — Sun, 03 Jun 2018 19:30:26 GMT

Below is my code. However, SAS Output has no information for HR and its 95%CI. What am I doing wrong here? Thanks in advance.

Hazard Ratio	95% Hazard Ratio Confidence Limits

proc import 
datafile="...\interaction.csv" out=cruise
dbms=csv replace;
getnames=yes;
run;

proc phreg data=help(where=(chemo ne 99 and stage ne 99));
class chemo (ref='5') stage (ref='2');            

model duration*vital_statusX(1)=chemo stage stage|chemo 
/rl selection=stepwise slentry=0.25 slstay=0.35 details;
run;

Re: Proc phreg /interaction between non-time covariates

Reeza — Sun, 03 Jun 2018 19:36:05 GMT

Use a HazardRatio Statement explicitly or RISKLIMITS on the MODEL statement.http://documentation.sas.com/?docsetId=statug&docsetTarget=statug_phreg_syntax17.htm&docsetVersion=14.3&locale=en#statug.phreg.phrmodrisklimits

Re: Proc phreg /interaction between non-time covariates

pau13rown — Sun, 03 Jun 2018 19:44:37 GMT

there are a number of things you could do eg

-before the phreg code add "ods output parameterestimates=hr (keep=HAZARDRATIO HRLOWERCL HRUPPERCL);" see here: https://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/viewer.htm#statug_phreg_sect041.htm,

-or use the hazardratio statement: https://support.sas.com/documentation/cdl/en/statug/63962/HTML/default/viewer.htm#statug_phreg_sect016.htm.

-or, it's possible that the interaction term falls out of the model because you are using a stepwise selection procedure, and thus it may not appear(?)

Re: Proc phreg /interaction between non-time covariates

Cruise — Sun, 03 Jun 2018 19:58:45 GMT

trying your suggestions one at a time. Removing stepwise selection didn't lead to produce hazard ratio estimates. Now getting into the links to study.

Re: Proc phreg /interaction between non-time covariates

Reeza — Sun, 03 Jun 2018 20:07:24 GMT

I downloaded your file and ran your code exactly as shown and got HR and confidence limits. You noticed that you're importing data and calling it CRUISE but then using a dataset named HELP in your code?

@Cruise wrote:
trying your suggestions one at a time. Removing stepwise selection didn't lead to produce hazard ratio estimates. Now getting into the links to study.

Re: Proc phreg /interaction between non-time covariates

Reeza — Sun, 03 Jun 2018 20:14:20 GMT

My bad PROC PHREG does not calculate the HR because it's an interaction effect. This means that to consider chemo or stage if the interaction effect is significant you essentially need to set a level for each to get the hazard ratio. So when you have interaction terms you MUST use a HAZARDRATIO statement with the AT to calculate the HR.

The earlier models have the HR but the last does not have the HR calculated. The table shows all the values as missing. My interpretation from your question was that no hazard ratio table was being presented. One is shown but has no values which is a different issue.

Re: Proc phreg /interaction between non-time covariates

Cruise — Sun, 03 Jun 2018 20:47:48 GMT

Now I know hazardratio statement is required, I'm following this SUGi paper.
https://support.sas.com/resources/papers/proceedings10/253-2010.pdf

Re: Proc phreg /interaction between non-time covariates

Cruise — Sun, 03 Jun 2018 21:35:15 GMT

Hi,

Code below gave me Hazard Ratios for different combinations of chemos at different "stages". However, Hazard Ratio Estimates has all the values missing in the first table for "Analysis of Max Likelihood Estimates". Any idea how to calculate HRs for non-interaction terms as well @Reeza @pau13rown

proc phreg data=actualdata;
class chemo(ref='5') stage(ref='2') / param=ref order=internal;
model duration*vital_statusX(1)=chemo stage chemo*stage/ties=Efron;

hazardratio 'chemo comparisons at stage=1' chemo / diff=ref at (stage='1');
hazardratio 'chemo comparisons at stage=2' chemo / diff=ref at (stage='2');
hazardratio 'chemo comparisons at stage=3' chemo / diff=ref at (stage='3');
hazardratio 'chemo comparisons at stage=4' chemo / diff=ref at (stage='4');
run;

SAS Output

Analysis of Maximum Likelihood Estimates
Parameter			DF	Parameter Estimate	Standard Error	Chi-Square	Pr > ChiSq	Hazard Ratio	Label
chemo	1		1	-0.58488	0.06679	76.6870	<.0001	.	chemo 1
chemo	2		1	-0.76999	0.06946	122.8931	<.0001	.	chemo 2
chemo	3		1	-0.36383	0.07624	22.7723	<.0001	.	chemo 3
chemo	4		1	-0.56654	0.04868	135.4679	<.0001	.	chemo 4
chemo	99		1	-0.55819	0.06683	69.7661	<.0001	.	chemo 99
stage	1		1	0.61075	0.21385	8.1564	0.0043	.	stage 1
stage	3		1	0.55288	0.05532	99.9003	<.0001	.	stage 3
stage	4		1	2.07708	0.05469	1442.4147	<.0001	.	stage 4
stage	99		1	1.46554	0.07563	375.4676	<.0001	.	stage 99
chemo*stage	1	1	1	0.15240	0.49794	0.0937	0.7596	.	chemo 1 * stage 1
chemo*stage	1	3	1	-0.24116	0.07703	9.8010	0.0017	.	chemo 1 * stage 3
chemo*stage	1	4	1	-0.43955	0.07787	31.8633	<.0001	.	chemo 1 * stage 4
chemo*stage	1	99	1	-0.34137	0.13388	6.5018	0.0108	.	chemo 1 * stage 99
chemo*stage	2	1	1	0.38167	0.43722	0.7620	0.3827	.	chemo 2 * stage 1
chemo*stage	2	3	1	-0.30036	0.07728	15.1070	0.0001	.	chemo 2 * stage 3
chemo*stage	2	4	1	-0.39476	0.07560	27.2679	<.0001	.	chemo 2 * stage 4
chemo*stage	2	99	1	-0.10439	0.12572	0.6895	0.4063	.	chemo 2 * stage 99
chemo*stage	3	1	1	0.11394	0.41748	0.0745	0.7849	.	chemo 3 * stage 1
chemo*stage	3	3	1	-0.31936	0.08747	13.3311	0.0003	.	chemo 3 * stage 3
chemo*stage	3	4	1	-0.64530	0.08719	54.7815	<.0001	.	chemo 3 * stage 4
chemo*stage	3	99	1	-0.65144	0.13323	23.9088	<.0001	.	chemo 3 * stage 99
chemo*stage	4	1	1	-0.91324	0.21560	17.9426	<.0001	.	chemo 4 * stage 1
chemo*stage	4	3	1	0.24049	0.05769	17.3802	<.0001	.	chemo 4 * stage 3
chemo*stage	4	4	1	0.32022	0.05720	31.3412	<.0001	.	chemo 4 * stage 4
chemo*stage	4	99	1	-0.17206	0.07879	4.7694	0.0290	.	chemo 4 * stage 99
chemo*stage	99	1	1	-0.95027	0.24051	15.6106	<.0001	.	chemo 99 * stage 1
chemo*stage	99	3	1	0.03088	0.08554	0.1303	0.7181	.	chemo 99 * stage 3
chemo*stage	99	4	1	-0.06898	0.08721	0.6256	0.4290	.	chemo 99 * stage 4
chemo*stage	99	99	1	-1.37108	0.10444	172.3358	<.0001	.	chemo 99 * stage 99

chemo comparisons at stage=1: Hazard Ratios for chemo
Description	Point Estimate	95% Wald Confidence Limits
chemo 1 vs 5 At stage=1	0.649	0.247	1.707
chemo 2 vs 5 At stage=1	0.678	0.291	1.581
chemo 3 vs 5 At stage=1	0.779	0.348	1.741
chemo 4 vs 5 At stage=1	0.228	0.151	0.344
chemo 99 vs 5 At stage=1	0.221	0.141	0.348

chemo comparisons at stage=2: Hazard Ratios for chemo
Description	Point Estimate	95% Wald Confidence Limits
chemo 1 vs 5 At stage=2	0.557	0.489	0.635
chemo 2 vs 5 At stage=2	0.463	0.404	0.531
chemo 3 vs 5 At stage=2	0.695	0.599	0.807
chemo 4 vs 5 At stage=2	0.567	0.516	0.624
chemo 99 vs 5 At stage=2	0.572	0.502	0.652

chemo comparisons at stage=3: Hazard Ratios for chemo
Description	Point Estimate	95% Wald Confidence Limits
chemo 1 vs 5 At stage=3	0.438	0.406	0.472
chemo 2 vs 5 At stage=3	0.343	0.321	0.366
chemo 3 vs 5 At stage=3	0.505	0.464	0.549
chemo 4 vs 5 At stage=3	0.722	0.679	0.767
chemo 99 vs 5 At stage=3	0.590	0.532	0.655

chemo comparisons at stage=4: Hazard Ratios for chemo
Description	Point Estimate	95% Wald Confidence Limits
chemo 1 vs 5 At stage=4	0.359	0.332	0.388
chemo 2 vs 5 At stage=4	0.312	0.294	0.331
chemo 3 vs 5 At stage=4	0.365	0.336	0.396
chemo 4 vs 5 At stage=4	0.782	0.737	0.829
chemo 99 vs 5 At stage=4	0.534	0.479	0.596

Re: Proc phreg /interaction between non-time covariates

Reeza — Sun, 03 Jun 2018 21:41:47 GMT

You cannot get a individual estimate for chemo or stage because they interact so you have to use hazard ratio statements as well for any other HRs you want.

Re: Proc phreg /interaction between non-time covariates

Cruise — Sun, 03 Jun 2018 21:57:04 GMT

@Reeza

adding "hazardratio chemo" to my model results in with added exhausting table below. Please correct in my code directly. I know it's spoon feeding but I'm really panicking to be running out of time. I appreciate your help Reeza! All I need is hazard ratios with CI for each level of chemo (1,2,3,4).

ods trace on;
proc phreg data=cmydata(where=(KRAS ne 99));
class chemo(ref='5') stage(ref='2') 
	  surgery (ref='5') radiation(ref='2')
	  hormone (ref='3') immuno(ref='3')
	  sex(ref='2')/param=ref order=internal ;

model duration*vital_statusX(1)=chemo stage chemo*stage surgery 
      radiation hormone immuno sex/ties=Efron rl;
hazardratio chemo; 
hazardratio 'chemo comparisons at stage=1' chemo / diff=ref at (stage='1');
hazardratio 'chemo comparisons at stage=2' chemo / diff=ref at (stage='2');
hazardratio 'chemo comparisons at stage=3' chemo / diff=ref at (stage='3');
hazardratio 'chemo comparisons at stage=4' chemo / diff=ref at (stage='4');
run;
ods trace off;

SAS Output

Hazard Ratios for chemo
Description	Point Estimate	95% Wald Confidence Limits
chemo 1 vs 2 At stage=1	5.497	1.486	20.332
chemo 1 vs 3 At stage=1	1.025	0.315	3.332
chemo 1 vs 4 At stage=1	1.269	0.000	8.36E235
chemo 1 vs 5 At stage=1	1.715	0.569	5.171
chemo 1 vs 99 At stage=1	3.340	0.419	26.589
chemo 2 vs 3 At stage=1	0.186	0.042	0.834
chemo 2 vs 4 At stage=1	0.231	0.000	1.52E235
chemo 2 vs 5 At stage=1	0.312	0.074	1.313
chemo 2 vs 99 At stage=1	0.607	0.063	5.881
chemo 3 vs 4 At stage=1	1.238	0.000	8.16E235
chemo 3 vs 5 At stage=1	1.673	0.446	6.273
chemo 3 vs 99 At stage=1	3.258	0.361	29.396
chemo 4 vs 5 At stage=1	1.351	0.000	8.89E235
chemo 4 vs 99 At stage=1	2.632	0.000	1.74E236
chemo 5 vs 99 At stage=1	1.948	0.226	16.748
chemo 1 vs 2 At stage=2	5.497	1.486	20.332
chemo 1 vs 3 At stage=2	1.025	0.315	3.332
chemo 1 vs 4 At stage=2	1.982	0.968	4.060
chemo 1 vs 5 At stage=2	1.715	0.569	5.171
chemo 1 vs 99 At stage=2	3.340	0.419	26.589
chemo 2 vs 3 At stage=2	0.186	0.042	0.834
chemo 2 vs 4 At stage=2	0.361	0.112	1.158
chemo 2 vs 5 At stage=2	0.312	0.074	1.313
chemo 2 vs 99 At stage=2	0.607	0.063	5.881
chemo 3 vs 4 At stage=2	1.934	0.697	5.370
chemo 3 vs 5 At stage=2	1.673	0.446	6.273
chemo 3 vs 99 At stage=2	3.258	0.361	29.396
chemo 4 vs 5 At stage=2	0.865	0.342	2.185
chemo 4 vs 99 At stage=2	1.685	0.231	12.291
chemo 5 vs 99 At stage=2	1.948	0.226	16.748
chemo 1 vs 2 At stage=3	1.198	0.860	1.669
chemo 1 vs 3 At stage=3	0.730	0.431	1.235
chemo 1 vs 4 At stage=3	0.907	0.626	1.313
chemo 1 vs 5 At stage=3	0.600	0.390	0.924
chemo 1 vs 99 At stage=3	0.792	0.456	1.375
chemo 2 vs 3 At stage=3	0.609	0.382	0.973
chemo 2 vs 4 At stage=3	0.757	0.573	1.001
chemo 2 vs 5 At stage=3	0.501	0.351	0.715
chemo 2 vs 99 At stage=3	0.661	0.403	1.085
chemo 3 vs 4 At stage=3	1.242	0.757	2.039
chemo 3 vs 5 At stage=3	0.822	0.478	1.414
chemo 3 vs 99 At stage=3	1.085	0.571	2.062
chemo 4 vs 5 At stage=3	0.662	0.448	0.977
chemo 4 vs 99 At stage=3	0.873	0.519	1.469
chemo 5 vs 99 At stage=3	1.319	0.750	2.321
chemo 1 vs 2 At stage=4	1.262	1.026	1.551
chemo 1 vs 3 At stage=4	0.878	0.629	1.225
chemo 1 vs 4 At stage=4	0.525	0.404	0.684
chemo 1 vs 5 At stage=4	0.512	0.367	0.714
chemo 1 vs 99 At stage=4	1.055	0.658	1.692
chemo 2 vs 3 At stage=4	0.696	0.527	0.918
chemo 2 vs 4 At stage=4	0.416	0.345	0.503
chemo 2 vs 5 At stage=4	0.406	0.308	0.535
chemo 2 vs 99 At stage=4	0.836	0.541	1.292
chemo 3 vs 4 At stage=4	0.598	0.434	0.825
chemo 3 vs 5 At stage=4	0.583	0.399	0.852
chemo 3 vs 99 At stage=4	1.202	0.727	1.986
chemo 4 vs 5 At stage=4	0.975	0.708	1.342
chemo 4 vs 99 At stage=4	2.009	1.262	3.197
chemo 5 vs 99 At stage=4	2.061	1.243	3.417
chemo 1 vs 2 At stage=99	0.833	0.086	8.074
chemo 1 vs 3 At stage=99	79333.75	0.000	1.11E202
chemo 1 vs 4 At stage=99	1.948	0.233	16.283
chemo 1 vs 5 At stage=99	0.900	0.081	9.986
chemo 1 vs 99 At stage=99	84109.03	0.000	5.17E246
chemo 2 vs 3 At stage=99	95197.47	0.000	1.32E202
chemo 2 vs 4 At stage=99	2.338	0.584	9.363
chemo 2 vs 5 At stage=99	1.080	0.180	6.465
chemo 2 vs 99 At stage=99	100927.6	0.000	6.19E246
chemo 3 vs 4 At stage=99	0.000	0.000	3.41E192
chemo 3 vs 5 At stage=99	0.000	0.000	1.58E192
chemo 3 vs 99 At stage=99	1.060	0.000	.
chemo 4 vs 5 At stage=99	0.462	0.093	2.290
chemo 4 vs 99 At stage=99	43171.78	0.000	2.65E246
chemo 5 vs 99 At stage=99	93473.51	0.000	5.74E246

Re: Proc phreg /interaction between non-time covariates

Reeza — Sun, 03 Jun 2018 21:58:51 GMT

Because you have an interaction term you cannot get a HR for CHEMO alone it's chemo at the different levels of stage which you already have above.